Abstract
Fossil fuels are the primary source of energy to global industrialization and rapid development and are consumed at an alarming rate. This creates a dire need to develop alternative techniques for analyzing the operating parameters of compression ignition (CI) diesel engines. The traditional approach of experimenting, simulating, and optimizing the CI engine operating parameters is complex, time-consuming, and costly. This has led the researchers to shift their focus to agile and computationally economic techniques like soft computing (SC) for predicting the optimum performance with a substantial accuracy. Artificial neural network (ANN), one of the SC techniques, has proven to be a viable alternative to traditional experimental and simulation methodologies. The present research work mainly focuses on the implementation of ANN for the prediction of performance, combustion and emission parameters of a single-cylinder, four-stroke, water-cooled, variable compression ratio CI engine powered by the conventional diesel fuel. The ANN models have been developed to predict the brake power, brake thermal efficiency, brake specific fuel consumption, ignition delay, combustion duration, carbon monoxide, carbon dioxide and oxides of nitrogen. These models have been trained by considering a small sized dataset acquired by experimenting on a single-cylinder CI diesel engine and using different training algorithms namely Levenberg-Marquardt, Scaled Conjugate Gradient and Broyden-Fletcher-Goldfarb-Shanno. All the ANN models considered are found to predict the operating parameters with a low root mean squared error (RMSE) and correlation coefficient (R) greater than 0.9, thereby implying a high resemblance between the experimental and the predicted values. The present study demonstrates the prediction capabilities of various ANN models using the small datasets.
